JP6928208B2 - Systems and methods for inverse transformation of chirp Z-transform with O (nlog (n)) time and O (n) memory - Google Patents

Description

The present invention generally relates to the inverse chirp z-transform, more specifically in O (n × log (n)) (hereinafter “O (nlog (n))”) time and O (n) memory. It relates to a method and a system for calculating the inverse chirp z-transform.

で与えられる（ここで、Ｗ_{イタリック}及びＡ_{イタリック}は、それぞれ、上記式においてイタリック体のＷ及びＡを示す）。 The Chapter Z-Transform (CZT) extends the Discrete Fourier Transform (DFT) by allowing samples to be placed on log-helical contours instead of unit circles. More specifically, the transformation _{distributes the sample along a log-helical contour defined by the formula A italic} W _italic- ^k (k = 0, 1, 2, ..., M-1). Complex A _italics and W _italic non-zero, the direction and the position of the spiral contour, as well as specify the interval between sample points on the contour line. More specifically, given the input vector x of the N element, CZT calculates the output vector X of the M element, where the kth element of X is

(Here, W _italic and A _italic represent italicized W and A in the above formula, respectively).

An efficient algorithm that can calculate the forward chirp z-transform has been described earlier. The algorithm operates in O (nlog (n)) time, where n is the size of the transformation. In the most common case, the computational complexity of the CZT algorithm is specified by n = max (M, N), as the number of inputs N and the number of outputs M can be different.

An efficient CZT algorithm can be derived using index substitution to represent the transform using fast convolution, which was originally "A liner filtering approach to the Computation of discrete Fourier transform". , IEEE Transactions on Audio and Algorithms, 18 (4): 451-455, 1970, proposed by Bluestein, which is incorporated herein by reference in its entirety. Various useful optimizations have been proposed for the CZT algorithm. Nevertheless, the computational complexity remains O (nlog (n)).

The inverse chirp Z-transform (ICZT) is the reverse of the chirp z-transform (CZT). That is, the ICZT maps the output of the CZT to the input in the opposite direction. Since CZT is a linear transformation, CZT can be expressed using the product of the matrix of the CZT matrix and the input vector. This matrix can be inverted using standard algorithms. In the algorithmic form, however, this process may ^{require up to O (n 3} ) operations.

Even if there is an efficient matrix inversion algorithm that runs faster than O (n ^{3 ), at least n 2} operations are required to access each element of the n × n matrix (n rows and n columns). be. Therefore, O (n ² ) is the lower limit of the computational complexity of the ICZT algorithm that works with the complete n × n matrix in memory.

To be efficient, the ICZT algorithm must have the same computational complexity as the CZT algorithm, i.e. O (nlog (n)). This requirement excludes any method that requires the transformation matrix to be stored in memory.

Several attempts have been made to derive an efficient ICZT algorithm. One case (Frickey, "Ussing the influence chirp-z transform for time-domain algorithm of simulated radar signs", Technical report, ideal. In the one described in (Incorporated), the modified version of the forward CZT algorithm is described as the ICZT algorithm, in which the logarithmic spiral contour is traversed in opposite directions. However, this method does not actually invert the CZT. This works only in some special cases, for example in the case of A _italic = 1 and W _italic = e- ^{2πi / n} . That is when the CZT results in a DFT. In the general case, i.e. A _italic , W _italic ∈ C \ {0}, this method produces a transformation that does not invert CZT.

The implementation of this application describes an efficient O (nlog (n)) method for calculating ICZT. This method uses a generated vector to represent a more compactly structured matrix in O (n) memory (eg, diagonal matrix, Van der Monde matrix, Toeplitz matrix, circular matrix, or skew circular matrix). do. The ICZT can be seen as a generalization of the inverse fast Fourier transform (IFFT) that deviates from the unit circle. In other words, the complex parameters A _italics and W _Italicized ICZT depict logarithmic spiral contour in the complex plane. Rather than like an IFFT, the ICZT corresponds to a frequency component that increases exponentially or decays exponentially and can use contour points that deviate from the unit circle. The implementation of the present application describes a method for improving the numerical accuracy of CZT or ICZT in the case of | W _italic | <1 ( _{the absolute value of "W italic} " is less than 1). Is given in Appendix A.

As far as they know, the inventors consider this application to be the first disclosure of an efficient ICZT algorithm. In particular, practical algorithms, and their derivations, are also explained. In addition, the accuracy of the algorithm is verified using automated test cases.

（以下、文章中で「Ｗ＾」と記す）を反転することに帰着する。 The algorithm was derived by expressing the CZT equation using structured matrix multiplication and then finding a way to reverse the equation. The essence of the ICZT calculation comes down to inversion of the specially constructed Vandermonde matrix W. This problem, in turn, is a specially constructed Toeplitz matrix derived from W.

It comes down to reversing (hereinafter referred to as "W ^" in the text).

Other aspects, objectives, and advantages of the present application will become more apparent from the following detailed description in conjunction with the accompanying drawings.

The accompanying drawings incorporated herein together and forming part of the specification show some aspects of the present application and, along with their description, serve to illustrate the principles of the present application. In the drawing:

FIG. 1A depicts an example of a logarithmic spiral contour with 32 points, where the start of the contour is indicated using an unfilled circle. FIG. 1B depicts an example of a logarithmic spiral contour with 64 points, where the start of the contour is indicated using an unfilled circle.

FIG. 2 shows the numerical precision for executing the CZT followed by the ICZT in the case of contour lines having shapes as shown in FIGS. 1A and 1B but having different points M and in the case of calculations using different floating point precisions. It is a chart which provides an example of.

FIG. 3 is a chart that provides examples of matrix shapes and generation vectors for different types of structured matrices.

FIG. 4A gives an example illustrating a fixed amplitude complex exponential function used by an FFT or IFFT. FIG. 4B gives an example illustrating an exponentially increasing or exponentially decaying complex exponential function that can be used by CZT or ICZT.

FIG. 5A shows an example of a logarithmic spiral contour line, the points of the contour line being located outside the reference circle. FIG. 5B shows an example of a logarithmic spiral contour line, the points of the contour line being placed inside the reference circle. FIG. 5C shows an example of a logarithmic spiral contour line, the points of the contour line starting from the outside of the reference circle and ending inside. FIG. 5D shows an example of a logarithmic spiral contour line, the points of the contour line starting from the inside of the reference circle and ending at the outside.

FIG. 6A shows an example of a logarithmic spiral contour line, the points of the contour line being placed on a unit circle over a range of 90 degrees. FIG. 6B shows an example of a logarithmic spiral contour line, the points of the contour line being placed on a unit circle over a range of 180 degrees. FIG. 6C shows an example of a logarithmic spiral contour line, the points of the contour line being placed on a unit circle over a range of 360 degrees.

FIG. 7A shows an example of a logarithmic spiral contour over a range of two 360-degree axial center rotations. FIG. 7B shows an example of a logarithmic spiral contour over a range of 5 axial center rotations of 360 degrees.

While the present application has been described in the context of certain preferred embodiments, it is not intended to be limited to those embodiments. On the contrary, its intent is to cover all alternatives, modifications, and equivalents contained within the spirit and scope of the present application as defined by the appended claims.

The embodiments of the present disclosure describe an efficient O (nlog (n)) ICZT algorithm that implements ICZT for _{A italics} , W _{italics ∈ C \ {0}.} The ICZT generalizes the Fast Inverse Fourier Transform (IFFT) by allowing sample points to be distributed on logarithmic spiral contours on the complex plane instead of unit circles. Two exemplary logarithmic spiral contours are shown in FIGS. 1A and 1B. Whenever the sample points are not on the unit circle, the corresponding frequency components increase or decrease exponentially. 4A and 4B show examples of increasing, damped, and constant amplitude frequency components used by the FFT / IFFT and CZT / ICZT.

The ICZT algorithm was derived by expressing the CZT equation as a product of structured matrices and finding a way to invert the matrix equation. More specifically, the calculation of ICZT results in reversing the special case of the Vandermonde matrix W. This can be achieved by reversing the special case of the Toeplitz matrix W ^ derived from W.

A structured matrix is a matrix that can be described with less than the total number of those elements. Examples of structured matrices include the Toeplitz matrix, the Hankel matrix, the Vandermonde matrix, and the Cauchy matrix. Other examples include a circulant matrix, a skew circulant matrix, and a DFT matrix. Diagonal matrices can also be seen as structured matrices. FIG. 3 shows some common shapes of these structured matrices. The example shown in FIG. 3 is the case of a 3 × 3 matrix. In addition, FIG. 3 illustrates a generation vector that can be used to generate each matrix. In most cases, they are represented by r and c, which specify the first row and first column of the matrix.

Gohberg and Semincul have shown that the inverse of the Toeplitz matrix can be represented using the difference between the two products of the Toeplitz matrix. This study is based on I.I. Gohberg et al. , "On the innovation of finish Toeplitz matrix and their contour analogs", Mat. issled, 2 (1): 201-233, 1972, and William Trench, "Algorithm for innovation of finance Toeplitz matrix", Journal of the Society Both of these references have been published and are incorporated herein by reference in their entirety. In the literature, this result is often referred to as the "Gohberg-Semencul formula". The four matrices in this equation are either upper triangular teplices or lower triangular teplices and are generated by two vectors u and v. In the case of ICZT, the Toeplitz matrix to be inverted is also symmetric. This leads to a simplified version of the Gohberg-Semencle equation that represents the inverse matrix using only one generation vector, i.e. only the vector u.

The present inventors have identified that in the case of ICZT, it is possible to derive an expression that defines the _{generation vector u as a function of complex number W italics.} This equation led to an efficient ICZT algorithm. This algorithm involves multiplying the Toeplitz matrix by a vector, which is done with O (nlog (n)), without storing the entire Toeplitz matrix in memory. Appendix C, D, and E implement three different algorithms based on the reference in which this product can be calculated. Each of these algorithms can be used as a building block for the ICZT algorithm. The Fast Fourier Transform (FFT) algorithm is described in Appendix B and is used in each of these three algorithms.

Wider impact.

Its efficient implementation using the Discrete Fourier Transform (DFT) and the Fast Fourier Transform (FFT) is deeply embedded in a wide variety of today's applications. Since CZT is a generalization of DFT and ICZT is a generalization of inverse DFT, the number of potential uses of the present application is very large. So far, only the CZT algorithm has the same computational complexity as the FFT, i.e. O (nlog (n)). The present application derives an ICZT algorithm with the same complexity as an inverse FFT (IFFT), which is also O (nlog (n)).

In other words, the application is mutable, not only because the application performs transformations that generalize the inverse FFT, but also because the new algorithm has the same execution time complexity as the algorithm that the application generalizes. Is.

A log spiral contour that can be used with the ICZT includes an outline curve, the points of the outline curve are outside the unit circle, inside the unit circle, and the outline curve starts outside the unit circle. It ends inside it, starts inside the unit circle and ends outside it. Examples of such contours are shown in FIGS. 5A-5D. Furthermore, the contour points may include partial arcs on the unit circle, such as 90 degree arcs, 180 degree arcs, or full 360 degree rotations, in which the contour lines are FFT / IFFT. Can be equivalent to the contour line used by. Examples of such contours are shown in FIGS. 6A-6C. Ultimately, contour points may be included in the subject even with multiple 360 degree rotations. 7A-7B show examples of two and five full rotations. In general, the number of rotations may not even be an integer.

Representing the chirp Z-transform using a structured matrix

Example of 4x4 CZT expressed in matrix format

This example describes a CZT for the special case with 4 inputs and 4 outputs using matrix notation. In this case, CZT is defined using the following equation (where W and A in italics in the equation indicate W _italics and A _italics in the text, respectively).

Suppose that x = (x ₀ , x ₁ , x ₂ , x ₃ ) T marks the CZT input vector, and X = (X ₀ , X ₁ , X ₂ , X ₃ ) ^T marks the output vector. .. Using these two vectors, a 4x4 CZT can also be expressed using the following matrix equation.

That is, the example of CZT can be seen as multiplying the input vector x by the diagonal matrix generated by the negative power of _{the complex parameter A italic, and multiplying the resulting vector by the Van der Monde matrix W.} The Mondo matrix W is specified by the _{complex parameter W italic.}

In this example, the matrix W has the following form:

Inverting the matrix W leads to ICZT.

Representing CZT using the Vandermonde matrix W

In the general case, CZT is typically defined using the following equation:

In this formulas, A _italics and W _italics are complex parameters for the conversion to define the position of the sample of the logarithmic spiral contour and logarithmic spiral contour. The parameter N is an integer that specifies the size of the input vector x. Finally, the parameter M is an integer that specifies the size of the output vector X. In general, N may not be equal to M. That is, the dimensions of the input do not have to be equal to the dimensions of the output.

CZT can also be represented in matrix form:

Equation (2.5) can be paraphrased in the following abbreviations:

Here, W is the following M × N matrix.

It can be seen that W is a Vandermonde procession. That is, each row of W forms a geometric progression. Furthermore, the common ratio of each progression of each of these sequences is equal to the power of the corresponding integer of the _{parameter W italics.}

_{The power of the negative integer of A italic in} equation (2.6) scales the column of the Vandermonde matrix W. Since the matrix W is a special case of the Vandermonde matrix, it can be expressed as the product of the diagonal matrix and the Toeplitz matrix W ^ as follows.

The Toeplitz matrix W ^ is derived from the Vandermonde matrix W.

Since the matrix W is a special case of the Vandermonde matrix, it can be expressed as the product of two diagonal matrices and the Toeplitz matrix W ^. The following equation can be used to represent the power _{of the parameter W italic in each element of the matrix W.}

Equation (2.8) means that the right-hand side of (2.4) can be expressed as follows.

A term in the last equation can be rearranged because it can be more easily converted to a matrix product:

The term W ^{k × k / 2} replaces the M × M diagonal matrix P. The term W ^{j × j / 2} A ^{−j replaces} the product of two N × N diagonal matrices Q and A. Finally, the term W ^{− (k−j) × (k−j) / 2A} −j is replaced by the M × N Toeplitz matrix W ^. W ^ is shown below.

この式において、すべての行列−ベクトルの積は、Ｏ（ｎｌｏｇ（ｎ））で計算されることができ、ここで、ｎ＝ｍａｘ（Ｍ、Ｎ）である。 It efficiently calculates the product of the matrix W ^ and the vector using one of several O (nlog (n)) algorithms that perform this operation (see Appendix C, D, and E). It means that you can do it. In other words, equation (2.10) has the following matrix form:

In this equation, the product of all matrices-vectors can be calculated by O (nlog (n)), where n = max (M, N).

ここで、Ｐ＝ｄｉａｇ（Ｗ_{イタリック} ^{０×０／２}、Ｗ_{イタリック} ^{１×１／２}、．．．、Ｗ_{イタリック} ^{（Ｍ−１）×（Ｍ−１）／２}）、Ｑ＝ｄｉａｇ（Ｗ_{イタリック} ^{０×０／２}、Ｗ_{イタリック} ^{１×１／２}、．．．、Ｗ_{イタリック} ^{（Ｎ−１）×（Ｎ−１）／２}）、Ａ＝（Ａ_{イタリック} ^−０、Ａ_{イタリック} ^−１、．．．Ａ_{イタリック} ^{−（Ｎ−１）}）であり、Ｗは、（２．７）において定義されたヴァンデルモンド行列の特殊なケースであり、Ｗ＾は、（２．１１）において定義されたテプリッツ行列の特殊なケースである。前に説明されたように、ｘは、ＣＺＴへの入力ベクトルであり、ＸはＣＺＴからの出力ベクトルである。 In summary, the CZT algorithm can be seen as an efficient implementation of the following matrix equations:

Here, P = diag (W _italic ^{0 ×} 0/2, W _italic ^{1 × 1/2} , ..., W _italic ^{(M-1) × (M-1) / 2} ), Q = diag (W _italic) ^{0 ×} _0/2 , W ^{italic 1 × 1/2} , ..., W _italic ^{(N-1) × (N-1) / 2} ), A = (A _italic- ⁰ , A _italic- ¹ ,. A _italic- ^(N-1) ), W is a special case of the Van der Monde matrix defined in (2.7), and W ^ is the Teplitz matrix defined in (2.11). This is a special case of. As explained earlier, x is the input vector to the CZT and X is the output vector from the CZT.

Represents the inverse of the Toeplitz matrix

Gohberg-Semencle equation for 4x4 cases

ここで、ａ、ｂ、ｃ、ｄ、ｆ、ｇ、及びｈは、行列Ｔを生成する７つの複素数である。 It is assumed that T is a 4 × 4 Toeplitz matrix. in short,

Here, a, b, c, d, f, g, and h are seven complex numbers that generate the matrix T.

ここで、ｕ＝（ｕ_０、ｕ_１、ｕ_２、ｕ_３）は、ｕ_０≠０を満たす４つの要素のベクトルであり、ｖ＝（ｖ_０、ｖ_１、ｖ_２、ｖ_３）は、ｖ_３≠０を満たす別の４つの要素のベクトルである。これら２つのベクトルは、行列Ｔを生成する数ａ、ｂ、ｃ、ｄ、ｆ、ｇ、及びｈによって特定される。しかしながら、Ｔを生成する７つの数の関数として明示的に表現することは困難である。 The Gohberg-Semencul equation makes it clear that the 4 × 4 non-singular Toeplitz matrix T can be inverted using the following equation.

Here, u = (u ₀ , u ₁ , u ₂ , u ₃ ) is _{a vector of four elements satisfying u 0} ≠ 0, and v = (v ₀ , v ₁ , v ₂ , v ₃ ) is. , V ₃ ≠ 0 is a vector of another four elements. These two vectors are identified by the numbers a, b, c, d, f, g, and h that produce the matrix T. However, it is difficult to express it explicitly as a function of seven numbers that generate T.

；２）ベクトルｖの反転によって生成された上三角テプリッツ行列、

；３）ｖを右へ一要素分だけシフトすることによって得られたベクトル（０、ｖ_０、ｖ_１、ｖ_２）によって生成された下三角テプリッツ行列

；４）ｕの反転を右へ一要素分だけシフトすることによって得られたベクトル（０、ｕ_３、ｕ_２、ｕ_１）によって生成された上三角形テプリッツ行列

を表す。 In summary, equation (3.2) is the inverse of a 4 × 4 Toeplitz matrix T using four structured matrices:
1) Lower triangular Toeplitz matrix generated by vector u

2) Upper triangular Toeplitz matrix generated by inversion of vector v,

3) Lower triangular Toeplitz matrix generated _{by the vector (0, v 0} , v ₁ , v ₂ ) obtained by shifting v to the right by one element.

4) Upper triangle Toeplitz matrix generated _{by the vector (0, u 3} , u ₂ , u ₁ ) obtained by shifting the inversion of u to the right by one element.

Represents.

A general case of the Gohberg-Semencul equation

Suppose T is a vector r = (r ₀ , r ₁ , r ₂ , ..., r _n-1 ) that specifies the first row and a vector c = (c ₀ , c ₁ , c) that specifies the first column. _2. It is assumed that it is an n × n Toeplitz matrix generated by (c _n-1). It is assumed that r ₀ = c _0. More formally

The inverse matrix T ^-1 of the Toeplitz matrix T may not be the Toeplitz. Nevertheless, the Gohberg-Semencle equation makes it possible to represent ^{T -1 using the upper and lower triangular Toeplitz matrices.}

That is, the Gohberg-Semencle equation has the vectors u = (u ₀ , u ₁ , u ₂ , ..., _un-1 ) and v = (v ₀ , v ₁ , v _{2) that satisfy the following matrix equations.} , ..., v _n-1 ) is specified to exist.

In other words, the inverse matrix T ^-1 can be regarded as a structured matrix generated by the vectors u and v. This analogy ^{extends to multiplying T -1} by a vector. When the generated vectors u and v are specified, then ^{the product of the matrix T-1} and the vector is O (nlog (n)) by performing (3.4) using the multiplication of the structured matrix. Can be calculated with.

Representing ICZT using a structured matrix

Equation (2.13) shows that CZT can be derived from a series of matrix-vector multiplications in which all matrices are diagonal except W ^, where W ^ is the Toeplitz matrix below.

This means that the ICZT can be considered as the product of the diagonal matrix, the inverse matrix W ^ ^-1 , and the vectors (see equation (2.13)).

Suppose u and v are two vectors that generate ^{W ^ -1} using the Gohberg-Semencle equation, that is.

Since the matrix W ^ is symmetric, its inverse matrix W ^ ^-1 is also symmetric. The formula for W ^ ^-1 is a special case of the Gohberg-Semencle formula. ^Examining the special case of the inverse matrix W ^ -1 for the first few values of n suggested three useful assumptions. First, the vector u is equal to the first column of ^{W ^ -1.} More formally

Second, the vector v is equal to the last column of ^{W-1, i.e.}

Third, the vector v is equal to the inversion of the vector u.

In other words, it is necessary to specify only the vector u so that the inverse matrix W ^ ^-1 can be multiplied by the vector using structured matrix multiplication.

One approach to deriving an explicit formula for computing the vector u is to derive an explicit formula from the special case for the first few values of n. If this induction is successful, then it should be possible to use the resulting formula for every finite n. This is sufficient to formulate the ICZT algorithm using Gohberg-Semencle equations and structured matrix multiplication, as u specifies v.

A special case of formulation for inverting the Toeplitz matrix W ^

Special case when n = 1

It is assumed that n = 1. Then W ^ is a 1x1 matrix consisting of a single number equal to 1. More formally

This means that the inverse matrix W- ¹ is also a 1 × 1 matrix consisting of a single numerical value 1. I'm gonna

The first and last lines of W ^ ^-1 are identical and consist of a single element, both equal to 1. More formally

This results in a special case of degeneracy of the Gohberg-Semencle equation:

Special case when n = 2

これは、Ｗ_{イタリック}∈Ｃ＼｛０｝毎に明確に定義される。 It is assumed that n = 2. Then W ^ is a 2 × 2 matrix specified by the following equation:

This is _{clearly defined for each W italic} ∈ C \ {0}.

In this case, the inverse matrix W ^ ^-1 is also a 2 × 2 matrix, which means that the first column u is a vector of two elements, that is, u = (u ₀ , u ₁ ) ^T. .. Its elements can be expressed as a function of _{W italics} by solving the following linear system.

In this equation, e ₀ represents a vector in which the first element is set to 1 and the remaining elements are set to zero, that is,

In other words, the dot product of the first row of W ^ and the vector u must be equal to 1, and the dot product of the second row of W ^ and the vector u must be equal to 0.

The linear system (4.13) can be represented in terms of the four elements of W ^ and the two elements of u. This leads to the next extended form (4.13).

The previous equation can also be seen as an equation for two scalar linear systems.

Both of these expressions are clearly defined whenever _{W italics ≠ 0.}

The term u ₀ W- ^{1 / 2} can be subtracted from both sides of (4.17), which leads to the equation for the value of _{u 1} as a function of _{W italics} and u _0:

The right-hand side of the above equation can be applied to (4.16). Thus, as a result, an expression that depends only on u ₀ and W _italics.

Multiplying both sides of the previous equation by W _italic / (W _italic- 1) yields an expression for the value _{u 0} as a function of _{W italic:}

Applying this value to the right side of (4.18) results in an expression representing the value _{of u 1} as a function of _{W italics.} More formally

In summary, in the 2x2 case, the column vector u that specifies the first column of the ^{inverse matrix W ^ -1} can be represented as the following function of the _{parameter W italics.}

The vector v can be obtained by reversing the elements of u.
in short,

The closed form representation for the inverse matrix W ^ ^-1 can be derived by combining the equations for u and v. in short,

In this special case, the Gohberg-Semencle equation can be verified using the following series of matrix transformations. The verification starts with the difference between the left and right sides of the equation described in (3.4) and shows that this difference is equal to zero.

Special case when n = 3

In this special case, W ^ is a 3x3 Toeplitz matrix. Each element of W ^ is a function of the _{conversion parameter W italic.} More formally

The inverse matrix W ^ ^-1 is also a 3x3 matrix, but unlike the 2x2 case, it is no longer a teplice. Suppose u is the first column of ^{W ^ -1, that is}

Similar to the 2x2 case, the vector u is the solution of the following linear system:

In contrast to the previous case, however, in this case e ₀ indicates a column vector consisting of three elements instead of two. The first element of e ₀ is equal to 1. The remaining two elements of e _{0 are equal to 0.} In other words,

The linear system (4.28) can be represented in terms of the nine elements that make up W ^ and the three elements that make up u. The results of this extension are shown below.

This linear system can also be expressed in scalar form using a system of three equations with three unknowns. The left side of each of these three equations is equal to the dot product between the row of W ^ and the vector u. The right-hand side is equal to the corresponding element of _{e 0.} More formally

Solving this system of linear equations leads to an equation representing the vector u as a function of the _{parameter W italics, which is shown below.}

The inverse matrix W ^ ^-1 has the following form.

Note that the vector v ^{specifies the last column of W ^ -1} and that column is equal to the inverse of the vector u, and the vector u specifies the first column of the matrix. In other words, the previous equation supports three assumptions (4.4), (4.5), and (4.6).

By repeating this process for other values of n, it becomes possible to derive a general formula that specifies u. The next section describes this formula.

General formula for inverting the Toeplitz matrix W ^

In the general case, the vector u ^{specifies the first column of W ^ -1} , and the elements of the vector u can be calculated using the following equation.

For each _k, the value of uk can be calculated in O (1) using the precomputed value of the product of the polynomials that appear in the denominator on the right side of (4.36). These values are calculated in separate loops, which operate in O (n).

The result value of u and its reversal can be applied to the Gohberg-Semencle equation (3.4) to calculate the ^{inverse matrix W ^ -1, and the reversal is equal to v.} It can also be used as a generated vector for an efficient algorithm that computes the product of a Toeplitz matrix and a vector. These algorithms can be called in a sequence that performs a vector multiplication of the matrix W ^ ^{-1 without actually storing W ^ -1 in memory.} This results in an ICZT algorithm that operates on O (nlog (n)).

ICZT in matrix format

ここで、Ａ^−１＝ｄｉａｇ（Ａ^０、Ａ^１、．．．、Ａ^Ｎ−１）、Ｑ^−１＝ｄｉａｇ（Ｗ^{０×０／２}、Ｗ^{１×１／２}、．．．、Ｗ^{（Ｎ−１）×（Ｎ−１）／２}）、及びＰ^−１＝ｄｉａｇ（Ｗ^{０×０／２}、Ｗ^{１×１／２}、．．．、Ｗ^{（Ｍ−１）×（Ｍ−１）／２}）である。 The following equation shows ICZT in matrix format:

Here, A ^-1 = diag (A ⁰ , A ¹ , ..., ^AN-1 ), Q ^-1 = diag (W ^{0 ×} 0/2, W ^{1 × 1/2} , ..., W ^{(N-1) x (N-1) / 2} ), and P- ¹ = diag (W ^{0 x} 0/2, W ^{1 x 1/2} , ..., W ^{(M-1) x (M-) 1) / 2} ).

That is, the equation shows how to calculate the input vector x of CZT from its output vector X. This equation was derived by inverting the diagonal matrix of the forward transformation equation (2.13) and replacing the ^{matrix W ^ with its inverse W ^ -1.}

Each matrix in equation (4.37) can be a diagonal matrix or expressed as a product difference of Toeplitz matrices. Performing the equation leads to an O (nlog (n)) algorithm for calculating the ICZT, which is given below in Algorithm 5.2. The algorithm efficiently performs all matrix calculations using the generated vector in order to calculate the results of all matrix-vector operations without storing the intermediate matrix in memory. The algorithm requires O (n) memory to operate.

example

In the 3x3 case, CZT can be expressed using the following matrix equations:

In that case, the extended matrix equation for ICZT looks like this:

ここで、ベクトルｕ＝（ｕ_０、ｕ_１、ｕ_２）は（４．３４）によって与えられ、ベクトルｖ＝（ｖ_０、ｖ_１、ｖ_２）＝（ｕ_２、ｕ_１、ｕ_０）はｕの逆である。ｕ及びｖのこれらの値に対して、式（４．４０）は、以下の形式を有する：

The inverse matrix W ^ ^-1 is given in (4.35). The algorithm, however, does not explicitly construct this matrix. As mentioned above, W ^ ^-1 can be expressed as follows.

Here, the vector u = (u ₀ , u ₁ , u ₂ ) is given by (4.34), and the vector v = (v ₀ , v ₁ , v ₂ ) = (u ₂ , u ₁ , u ₀ ). Is the opposite of u. For these values of u and v, equation (4.40) has the following form:

は、行列

の転置に等しく、また行列

は、行列

の転置に等しい。これ故に、逆行列W＾^−１は、以下の簡単な形式

で表される。さらには、

及び

の両方は、ベクトルｕの要素から生成されることができる。 In addition, the matrix

Is a matrix

Equal to the transpose of, and also the matrix

Is a matrix

Is equivalent to the transpose of. Therefore, the inverse matrix W ^ ^-1 has the following simple form:

It is represented by. Moreover,

as well as

Both can be generated from the elements of the vector u.

代わりに、式（４．４３）は、これらの行列の構造をうまく活用することによって効率的に計算される。上記のように、これら４つのテプリッツ行列は明示的に構築されていない。 Substituting (4.42) into (4.37) yields the following matrix equation for ICZT:

Instead, equation (4.43) is calculated efficiently by making good use of the structure of these matrices. As mentioned above, these four Toeplitz matrices are not explicitly constructed.

CZT algorithm and ICZT algorithm

Algorithm 5.1 provides pseudo-code for the CZT algorithm. The algorithm uses the convolutional algorithm TOEPLITZMULTIPLYC to multiply the Toeplitz matrix by a vector, which is described in Appendix E. Other algorithms for calculating the product of the Toeplitz matrix and the vector can be used instead. For example, the TOEPLITZMULTIPLYE described in Appendix C and the TOEPLITZMULTIPLYP described in Appendix D can be replaced by the TOEPLITZMULTIPLYC in Algorithm 5.1 without changing the computational complexity.

Algorithm 5.2 provides pseudo code for the ICZT algorithm. This implements the inverse matrix equation (4.42) without actually storing the matrix in memory. The algorithm operates on O (nlog (n)), where n = max (M, N). This implementation only supports the case n = M = N. As mentioned above, the function TOEPLITZMULTIPLYC is used on lines 23-26, and this function can be replaced by TOEPLITZMULTIPLYE or TOEPLITZMULTIPLYP without changing the computational complexity of the algorithm.

を用いて特定された。数値精度は、ＩＣＺＴが続くＣＺＴを実行すると共にＣＺＴの入力とＩＣＺＴの出力との間のユークリッド距離を計算することによって計算された。最終的な精度は、１００個のランダムな入力ベクトルにわたって平均された。図２の第２列は、変換行列の条件数を示す。高い条件数にもかかわらず、この問題は、高精度の浮動小数点数が使用されている場合に、Ｍの大きい値でも可解である。 FIG. 2 gives the numerical precision of ICZT as a function of the number of bits used by the floating point number during the calculation and the conversion size M. The two logarithmic spiral contours in FIGS. 1A and 1B correspond to M = 32 and M = 64 in FIG. The A _italic value was set to 1.1 for all rows in FIG. The value of W _{italic is the formula}

Was identified using. Numerical accuracy was calculated by executing CZT followed by ICZT and calculating the Euclidean distance between the input of CZT and the output of ICZT. The final accuracy was averaged over 100 random input vectors. The second column of FIG. 2 shows the conditional number of the transformation matrix. Despite the high conditional number, this problem is solvable even for large values of M when precision floating point numbers are used.

Algorithm 5.3 provides pseudo-code for the ICZT-RECT algorithm, which implements the inverse algorithm in the case of N ≦ M. This algorithm calls algorithm 5.2 with the second parameter set to M and then returns only the first N elements of its output vector.

Having described the ICZT algorithm, applications for the algorithm are now provided here. In that regard, the following is only a partial list of some useful applications of the ICZT algorithm. In an embodiment, the ICZT algorithm is used in applications for analyzing or generating data received from or transmitted to a radar sensor. In another embodiment, the ICZT algorithm is used in applications that generate or analyze data transmitted to or received from a sonar sensor. In yet another practice, the ICZT algorithm is used in applications for generating or analyzing data transmitted from or received by other ranging sensors.

In yet another embodiment, the ICZT algorithm is used in the hardware implementation of the ICZT algorithm. In an embodiment, the hardware implementation includes various systems including memory devices and processors, such as personal computers, chips (eg, system-on-chip), systems including application-specific integrated circuits (ASICs), and graphics processing units (such as graphics processing units). A system that includes a GPU), or a field programmable gate array (FPGA).

In further implementation, the ICZT algorithm improves the speed of applications that require computing ICZT by solving a slower method of solving an explicitly formulated linear system or explicitly inversion of a matrix. Used when doing. In addition, the ICZT algorithm can replace the conventionally used FFTs and IFFTs in whole or in part. This permutation can extend the scope of the computational system to include points on the log-helical contour inside or outside the unit circle. Furthermore, prior to or during the calculation of the ICZT, the frequency domain vector is modified so that the time domain vector returned by the ICZT is different from the time domain vector used to generate the frequency domain vector. Can be done. For example, an operation can be performed on a frequency domain vector, such as filtering a particular element or frequency from the vector.

In addition, in the examples, the ICZT algorithm is used in medical imaging applications such as CT, PET, or MRI. In addition, the ICZT algorithm is used in applications in signal processing, signal analysis, signal synthesis, speech and speech processing, and image processing. For signal processing applications, the CZT or FFT input vector can be a time domain vector derived from an audio or image signal, such as sampling a signal. The output vectors of the CZT or FFT may then be frequency domain vector representations of those signals.

The following Appendix A-E provide additional information regarding the implementation of the ICZT algorithm.

Appendix A: Inverting the direction of the log-helical contour to improve the numerical accuracy of CZT and ICZT.

This appendix describes alternative versions of the CZT and ICZT algorithms that improve the numerical accuracy of the calculated transformations. This is to invert the direction of the chirp outline when _{| W italic} | ( _{absolute value of W italic} ) <1 and to keep the original direction when _{| W italic} | ( _{absolute value of W italic) ≥ 1.} Achieved by. The parameters for the inverted contours are W _italic '= W _italic- ¹ and A _italic '= A _italic W _italic- ^(1-M) . Depending on the specific value of the conversion parameter, the improvement in numerical accuracy can exceed several orders of magnitude.

Algorithm A. 1 gives pseudocode for an alternative version of the CZT algorithm that performs contour line inversion when | W _italic | ( _{absolute value of W italic) <1.} Algorithm A. 2 provides pseudo-code for an alternative version of the ICZT algorithm that also performs contour inversion in that case.

Using these algorithms, the numerical accuracy of both CZT and ICZT can be improved for an _{increasing logarithmic chirp contour, i.e. | W italic | <1.} In those cases, Algorithm A. 1 and algorithm A. 2 reverses the direction of the chirp contour line. Furthermore, this inversion occurs with the inversion of the order of the elements in the vector X, which is described in Algorithm A. CZT output vector and algorithm in 1. It is an ICZT input vector in 2.

Appendix B: FFT and inverse FFT algorithms

This appendix describes the FFT algorithm and the inverse FFT algorithm. Algorithm B. 1 provides a pseudo code for the Fast Fourier Transform (FFT). Algorithm B. 2 provides pseudo code for the inverse fast Fourier transform (IFFT). Both FFT and IFFT operate at O (nlog (n)).

Appendix C: Multiplying the Toeplitz matrix by a vector

This appendix describes a general method of multiplying a Toeplitz matrix by a vector. Let A be an n × n Toeplitz matrix and x be a vector. The matrix A is specified by the first row r and the first column c, where r ₀ is assumed to be equal to _{c 0.} In the examples in this appendix, it is assumed that n is a power of 2. This approach, however, can be modified to work for other values of n as well. More formally

（以下、文章中で「Ａ＾」と記す）に埋め込むことが可能である：

In order to efficiently calculate the product y = Ax using the FFT algorithm, A is a 2n × 2n circulant matrix as shown below.

It can be embedded in (hereinafter referred to as "A ^" in the text):

（以下、文章中において「ｘ＾」と記す）になる：

The vector x is finally filled with n zeros, resulting in a vector of length 2n.

(Hereafter, it will be referred to as "x ^" in the text):

（以下、文章中で「ｙ＾」と記す）を計算することであり、そのベクトウは、Ａ＾ｘ＾に等しい。より正式には、

である。 The product of A ^ and x ^ can be calculated with O (nlog (n)) using an FFT algorithm. That is, the goal is a vector with a length of 2n.

(Hereinafter, it is referred to as "y ^" in the text), and its vector is equal to A ^ x ^. More formally

Is.

The first n elements of y ^ are equal to the corresponding elements of y, that is, each k ∈ {0, 1, 2, ... .. .. , N-1}, y _k = y ^ _k .

と置く。 Only the method of multiplying the (2n) × (2n) circulant matrix B by the vector v of length 2n using FFT and IFFT remains. The matrix B is generated by the first column vector b.
More formally

And put.

ここで、×は要素ごとの乗算を表す。従って、

である。 This product can be calculated using the DFT and the inverse DFT, which is obtained from the circular convolution theorem. In other words, each element (Bv) _k of the product is equal to the kth element in the circular convolution of b and v. in short,

Here, × represents multiplication for each element. Therefore,

Is.

DFT and inverse DFT can be calculated using FFT and inverse FFT, i.e.

In summary, the product of the n × n Teplitz matrix and the vector, where n is assumed to be a power of 2, the product can be calculated using the following 6 steps: 1) First Proceed from Teplitz to the cycle by concatenating columns, a single zero, and the inversion of the last n-1 element of the first row; 2) fill the vector with n zeros at the end; 3) of the circular matrix Calculate the DFT in the first column; 4) Calculate the DFT of the embedded vector; 5) Multiply the two DFTs element by element; 6) Calculate the inverse DFT of the product of each element; and 7) Returns the first n elements of the result vector calculated by the inverse DFT. The FFT algorithm is used to calculate the DFT, and the IFFT algorithm is used to calculate the inverse DFT. It should be noted that in step 1) neither the Toeplitz matrix nor the circulant matrix is explicitly calculated. Instead, only the generated vector is used to perform the required calculations.

Algorithm C. 1 provides pseudo code for the above algorithm. The computational complexity of this algorithm is O (nlog (n)). The name of the function is TOEPLITZMULTIPLYE, where "E" is an abbreviation for "embedding".

Instead of including the FFT calculation in the algorithm, it is possible to use FCIRCULANTMULTIPLY (see Algorithm D.2 below). However, FCIRCULANTMULTIPLY does more work O (n) than the sequence of FFT calculations, because for every ^{k ∈ {0, 1, ..., N-1, (f) 1 / n} and (f k) ^{1 / n} . Because it has to be calculated.

Appendix D: Product of Teplice and Vector Using Pushylnikov Decomposition

The algorithm for multiplying the Toeplitz matrix by a vector can be formulated using the Pushylnikov decomposition. This algorithm does not embed the Toeplitz matrix in the circulant matrix, as described in Appendix C. Instead, this appendix describes different versions of TOEPLITZMULTIPLY.

An example exemplifying this algorithm is shown below. Suppose A is a 4x4 Toeplitz matrix, which is generated by the first row r and the first column c.

ここで、Ａ´は、循環行列であり、Ａ´´は、ｆ＝−１を持つｆ循環行列（つまり、スキュー循環行列）である。この例の場合では、これら２つの行列は、以下の形式を有する：

Next, A can be decomposed into the sum of the two matrices as follows.

Here, A ′ is a circular matrix, and A ′ ′ is an f circular matrix having f = -1 (that is, a skew circular matrix). In the case of this example, these two matrices have the following form:

Formulas for calculating the values of A'and A''are provided below. If _{a'= (a'0, a'1} , a'2, ..., a'n-1) is assumed to be the first column of A', i.e.,

Similarly, suppose that a ″ = (a ″ ₀ , a ″ ₁ , a ″ ₂ , ..., a ″ _n-1 ) is the first column of A ″. in short,

ここで、ｒはＡの最初の行であり、ｃはそれの最初の列である。 Next, the elements of a'and a''can be calculated using the following equations.

Where r is the first row of A and c is the first column of it.

Algorithm D. 1 gives a pseudo code for the above algorithm. The name of the function is TOEPLITZMULTIPLYP, where "P" is an abbreviation for "Pustylnikov". Lines 27 and 28 are described in Algorithm D. Calling 2, which is explained below. The algorithm operates in O (nlog (n)) time.

Algorithm D. 2 gives a pseudo code (for multiplying the f-circulant matrix by a vector) for the function fcyclulant multiply, which is described in Algorithm D. Used in l.
It works with O (nlog (n)).

Appendix E: Product of Teplice and Vector Using FFT Convolution

Algorithm E. Reference numeral 1 indicates a pseudo code for yet another algorithm for multiplying the Toeplitz matrix by a vector. The function name in this case is TOEPLITZMULTIPLYC, where "C" abbreviates convolution. The computational complexity for the algorithm is O (nlog (n)). This algorithm can be interpreted as a form of circular embedding and is similar to the approach described in Appendix C.

Algorithm E. 2 gives a pseudo code for an FFT-based convolution algorithm, which is an O (nlog (n)) method for calculating discrete convolutions. It is the algorithm E. Used in 1.

(One side)
In a method that extends the scope of the computational system to include the logarithmic spiral contours of equation A _italic W _italic- ^k _{, A italic} W _italic is a non-zero complex number, k = 0, 1, ..., M. It is -1. The computing system uses at least one of the Fast Fourier Transform (FFT) or the inverse FFT (IFFT) in one or more processes, and the method replaces the IFFT or FFT in the inverse chirp Z transform in at least one process. A step of utilizing (ICZT) can be provided.
In this method, the ICZT can be represented in terms of a Vandermonde matrix with dimensions of M × N, and the computational complexity of using the ICZT on the computational system is equal to ^{O (n 2).} Or less than, where n = max (M, N).
All references, including publications, patent applications, and patents cited herein, are specified herein as a whole, with each reference individually and specifically indicated and incorporated by reference. Incorporated by reference to the same extent as if it were.

The use of the terms "a" and "an" and "the" as well as similar referents is apparent in the context of the description of the present application (particularly in the claims below) by particular indication or context herein. Unless there is a contradiction, both singular and plural objects are included. The terms "comprising," "having," "inclusion," and "contining" are non-limiting terms (ie, "including, but limited to" unless otherwise noted. It should be interpreted as "not done"). The enumeration of the range of values herein is intended only to serve as a shorthand notation for individually referencing each individual value within the range, unless otherwise specified herein. Each of the values of is incorporated herein as if it were described individually herein. All methods described herein can be performed in any suitable order, as long as there is no apparent contradiction by any particular instruction or context herein. Any and all examples provided herein, or the use of exemplary terms (eg, "such as"), are intended solely to better describe the present application and, unless specifically requested, the scope of the present application. It does not impose any restrictions on. There is no word in the specification that should be construed as indicating an unclaimed element as essential to the realization of the present application.

Preferred embodiments are described herein and include the best embodiments known to the inventor to carry out the present application. Modifications of those preferred embodiments may be apparent to those skilled in the art in reading the above description. We expect those skilled in the art to use such modifications appropriately, and we provide inventions to be practiced other than those specifically described herein. Invented to. Accordingly, the present application includes all modifications and equivalents of the subject matter listed in the claims attached herein, as permitted by applicable law. Moreover, any combination of the above elements in all possible variations thereof is incorporated herein by the present application unless there is a clear contradiction in the specification or context.

ICZT inverse chirp Z-transform CZT chirp z-transform

Claims (54)

A method of inverting the chirp Z-transform (CZT) using a computational system.
The computing system comprises a processor and a memory device.
The CZT has a parameter A _italics and parameters W _italics, the parameter A _italics and the parameter W _Italics, k = 0, 1, 2, ..., the formula A _italics W _italics ^-k against M-1 Define a logarithmic italic in the complex plane through
The CZT can be expressed as X = WAx, where W is a Van der Monde matrix of dimension M × N defined by the _{parameter W italics and their powers, where A is the parameter A italics} and It is the first diagonal matrix defined by its power,
The x is the first vector, and the k-th element of the X is k = 0, 1, 2, ..., M-1.

Is a second vector calculated from the first vector x using the CZT (where W _{italics and A italics indicate} W and A in italics, respectively). ,
The parameter A _italic and the parameter W _italic are non-zero complex numbers, and the X is stored in the memory device.
The method is:
A step in which the matrix W is represented as the product of the second diagonal matrix P, the Toeplitz matrix W ^, and the third diagonal matrix Q, wherein the CZT is represented as X = PW ^ QAx.
x = A ^-1 Q ^-1 W ^ ^-1 P ^-1 X is a step representing the inverse CZT, and A ^-1 , Q ^-1 , W ^ ^-1 , and P ^-1 are A and Q, respectively. , W ^, and the inverse matrix of P, step and
Calculate the inverse CZT by performing a right-to-left multiplication in the product of the ^{A -1} Q ^-1 W ^ ^-1 P ^-1 X to calculate the third vector x ^, using the processor. A method that includes steps to do. The method of claim 1, wherein each element of the first vector x is substantially equal to the corresponding element of the third vector x ^. The method of claim 1, wherein at least one element of the first vector x is not equal to the corresponding element of the third vector x ^. 3. The method of claim 3, further comprising performing an operation on the second vector X prior to the step of calculating, wherein the operation modifies at least one element of the second vector X. the method of. The method according to claim 4, wherein the calculation is a filter processing calculation. The method of claim 1, wherein the method has ^{computational complexity less than or equal to O (n 2} ) and n = max (M, N). The method of claim 6, wherein the computational complexity is O (nlog (n)). The matrix W ^ ^-1 is
W ^ ^-1 = (1 / u ₀ ) ×

Can be expressed as here

Is the first lower triangular Toeplitz matrix,

teeth,

The first upper triangular Toeplitz matrix, which is equivalent to the transpose of

Is the second upper triangular Toeplitz matrix,

teeth,

The second lower triangular Toeplitz matrix, which is equivalent to the transpose of
u ₀ is

Equal to
n is based on at least one of M or N,
The method according to claim 1. 4 said matrix

,

,

Or

The generated vector for each of

Can be derived from the vector u = (u ₀ , u ₁ , ..., un _{-1) given by,}
The method according to claim 8. Vector and W ^ ^-1 = (1 / u ₀ ) ×

8. The method of claim 8, wherein during the step of calculating the product of, the multiplication is performed from right to left so that no matrix is multiplied by another matrix. The Toeplitz matrix

,

,

Or

Multiplying at least one of the vectors by a vector further
A step of embedding the Toeplitz matrix in a circulant matrix , wherein the circulant matrix is represented by a generation vector of the circulant matrix, and the dimension of the circulant matrix is based on at least one of M and N.
A step of packing the vector with zeros prior to multiplication to generate a packed vector of the same length as the number of columns in the circulant matrix.
The step of calculating the product vector by multiplying the circulant matrix by the packed vector,
10. The method of claim 10, comprising extracting a result vector containing elements of the product vector corresponding to the row of the circular matrix in which the Toeplitz matrix is embedded. The method according to claim 11, wherein the circulant matrix is a square matrix, and the number of rows of the circulant matrix is a power of 2. The Toeplitz matrix

,

,

Or

Multiplication of a vector onto at least one of the above involves performing a Pushylnikov decomposition of the matrix prior to multiplication with the vector.
The method according to claim 10. The Toeplitz matrix

,

,

Or

Multiplying at least one of the vectors by a vector
A step of embedding the Toeplitz matrix in an extended Toeplitz matrix by packing the generated vector for the Toeplitz matrix with zeros, wherein the extended Toeplitz matrix is a dimension based on at least one of M or N. Have, step and
A step of packing zeros into the vector multiplied by the Toeplitz matrix to generate a packed vector having the same length as the number of columns in the extended Toeplitz matrix.
A step of expressing the extended Toeplitz matrix as the sum of a circulant matrix and a skew circulant matrix using the Pustylnikov decomposition.
A step of multiplying the circular matrix by the packed vector, multiplying the skew circular matrix by the packed vector, and adding the two result vectors to generate an embedded result vector.
A step of extracting a result vector from the packed result vector by retaining the elements of the packed result vector corresponding to the row of the extended Toeplitz matrix in which the Toeplitz matrix is embedded.
10. The method of claim 10. 14. The method of claim 14, wherein the number of rows and columns in the extended Toeplitz matrix is a power of two. During the execution of the method, the matrices A ^-1 , Q ^-1 ,

,

,

,

The method according to claim 8, wherein neither of the above and P- ^{1 is completely stored in the memory.} The method of claim 1, wherein no memory greater than O (n) is required to perform the method, where n = max (M, N). The method of claim 1, wherein M is equal to N. The method of claim 1, wherein M is not equal to N. The method according to claim 1, wherein the first vector x is derived from an audio signal. The method of claim 20, wherein the audio signal includes an audio signal. The method of claim 20, wherein the audio signal comprises at least one of a sonar signal or an ultrasonic signal. The method according to claim 1, wherein the first vector x is derived from an image signal. 23. The method of claim 23, wherein the image signal comprises at least one of a computed tomography (CT) signal, a positron emission tomography (PET) signal, or a magnetic resonance imaging (MRI) signal. The method according to claim 1, wherein the first vector x is derived from a signal received by a radar system sensor. The method of claim 1, wherein the first vector x is used to generate a signal transmitted to a radar sensor. The method of claim 1, wherein the second vector X is not calculated from a particular first vector x using the CZT having _{the same parameters A italics} and W _italics. A method of extending the scope of the computational system to include the logarithmic spiral contours of equation A _italic W _italic- ^k _{, where A italic} and W _italic are non-zero complex numbers and k = 0. 1, ..., M-1, the computational system uses at least one of a Fast Fourier Transform (FFT) or an inverse FFT (IFFT) in one or more processes.
The computing system comprises a processor and a memory, which calculates a frequency domain vector from the FFT or a time domain vector from the IFFT and stores the frequency domain vector and the time domain vector in the memory.
The method comprises the step of calculating the time domain vector using the processor by utilizing the inverse chirp Z-transform (ICZT) instead of the IFF T in at least one process.
The ICZT can be represented in terms of a Vandermonde matrix with dimensions M × N, and the computational complexity of using the ICZT on the computational system is less than or equal to ^{O (n 2).} Yes, where n = max (M, N), the method. 28. The method of claim 28, wherein the computational complexity is O (nlog (n)). 29. The method of claim 29, wherein the computing system uses only O (n) of memory during execution of the ICZT. A method of increasing the efficiency of a computational system that inverts the chirp Z-transform (CZT), which can be represented in terms of a Vandermonde matrix with dimensions M × N.
The computing system comprises a processor and memory, the method comprises the step of using a method of inverting the CZT to generate a time domain vector stored in the memory, the method performing on the processor. is a n = max (M, n) and is equal to or below the O ^{(n 2)} is the computational complexity at the time is, the CZT is represented by the parameter by a complex number a _italics and W _italicized two nonzero the complex a _italics and W _italic, k = 0, 1, ..., define a logarithmic spiral contour through the equation a _italics W _italics ^-k in M-1, method. 31. The method of claim 31, wherein the calculation system requires only O (n) of memory during the calculation of the inverse CZT. 31. The method of claim 31, wherein the method for reversing the CZT has computational complexity of O (nlog (n)). The numerical accuracy of the calculation system a method of the above improvement, the computing system, wherein using a processor of a computing system, a chirp-Z transform (CZT) or inverse chirp Z transform at least one of (ICZT), 1 or It calculates in a plurality of processes and stores the time domain vector or the frequency domain vector in the memory of the calculation system.
The CZT, the ICZT, or both the CZT and the ICZT have two non-zeros defining a logarithmic spiral contour _{in k = 0, 1, ..., M-1 via the formula A italic} W _italic- ^k. is it possible to represented by a parameter by the complex A _italics and W _italics,
In the case of | W _italic | <1, the method includes a step of reversing the direction of the log-helix contour so that the start and end points of the log-helix contour are exchanged in at least one process. ,
A method in which the order of elements in both the output vector of the CZT, the input vector of the ICZT, or the output vector of the CZT and the input vector of the ICZT is reversed. A method of converting a frequency domain vector X having a length M stored in the memory of a calculation system into a time domain vector x having a length N stored in the memory of the calculation system.
In order to calculate the time domain vector x, x = (1 / u ₀ ) A ^-1 Q ^-1

It comprises a step of calculating the inverse chirp Z-transform (ICZT), which can be represented as ^{P -1} X, by performing a right-to-left multiplication using the processor of the computing system.
A ^-1 is the first diagonal matrix defined by the parameter A _{italics and their powers.}
Q ^-1 is a second diagonal matrix based on the _{parameter W italic and its power.}

Is the first lower triangular Toeplitz matrix,

teeth,

The first upper triangular Toeplitz matrix, which is equivalent to the transpose of

Is the second upper triangular Toeplitz matrix,

teeth,

The second lower triangular Toeplitz matrix, which is equivalent to the transpose of
P ^-1 is a third diagonal matrix based on the _{parameter W italic and its power.}
The parameters A _italic and W _italic are non-zero complex numbers, which define a logarithmic spiral contour for k = 0, 1, ..., M-1 _{via the equation A italic} W _italic- ^k. death,
u ₀ is

Equally (where W _italic indicates italicized W),
n is a method based on at least one of M and N. The frequency domain vector X can be expressed as X = WAx ^, where W is _{a Vandermonde matrix defined by the complex parameter W italic} and its power, where A is the kth vector A. The elements of k = 0, 1, 2, ..., M-1

As given by, it is a fourth diagonal matrix based on the _{complex parameter A italic and its power (where A italic indicates} A in italics), where x ^ is the second time domain of length N. 35. The method of claim 35, which is a vector. The matrix W is composed of the product of the fifth diagonal matrix P, the Toeplitz matrix W ^, and the sixth diagonal matrix Q so that the frequency domain vector X is represented as a charp Z transform (CZT) in matrix form. Can be represented,
Here, the method according to claim 36, wherein X = PW ^ QAx. 35. The method of claim 35, wherein the frequency domain vector X is not calculated from the specific time domain vector x using _{the same parameters A italics} and W _italics using the chirp Z transform. 35. The method of claim 35, wherein only the number of O (n) is stored in memory during the execution of the method and n = max (M, N). Matrix A ^-1 , Q ^-1 ,

,

,

,

, And P- ^1, the method of claim 35, wherein none of the methods are completely stored in memory during execution of the method. 35. The method of claim 35, wherein the computational complexity of the method ^{is equal to or less than O (n 2} ), where n = max (M, N). The method of claim 41, wherein the computational complexity is O (nlog (n)). A memory device, wherein the memory device is configured to store a frequency domain vector X of length M.
A processor configured to map the frequency domain vector X to the time domain vector x using inverse chirp Z-transform (ICZT).
With
The ICZT is
x = (1 / u ₀ ) A ^-1 Q ^-1

^Represented as P-1 X,
Here, the processor performs matrix multiplication of the ICZT from right to left in order to calculate the time domain vector x of length N, where.
A ^-1 is the first diagonal matrix defined by the parameter A _{italics and their powers.}
Q ^-1 is a second diagonal matrix based on the _{parameter W italic and its power.}

Is the first lower triangular Toeplitz matrix,

teeth,

The first upper triangular Toeplitz matrix, which is equivalent to the transpose of

Is the second upper triangular Toeplitz matrix,

teeth,

The second lower triangular Toeplitz matrix, which is equivalent to the transpose of
P ^-1 is a third diagonal matrix based on the _{parameter W italic and its power.}
X is the frequency domain vector,
The parameters A _italic and W _italic are non-zero complex numbers that define a logarithmic spiral contour line for k = 0, 1, ..., M-1 _{via the equation A italic} W _italic- ^k.
u ₀ is

Equally (where W _{italic indicates italicized} W in the above equation),
n is a system based on at least one of M and N. The matrix

,

,

,as well as

The generated vector for each of

43. The system of claim 43, which can be derived from the vector u = (u ₀ , u ₁ , ..., un _{-1) given by.} Further comprising an input device for receiving a signal, the processor is coupled to said input device, said processor is configured to convert the signal time domain to the vector x ^, wherein the processor is Chi Sharp Z 43. The system of claim 43, which uses transformation to generate the frequency domain vector X from the time domain vector. The processor can be further configured to perform an operation on the frequency domain vector X before mapping the frequency domain vector X to the time domain vector x using the ICZT. 43. The system of claim 43, wherein at least one of the elements of the frequency domain vector X is modified. ^{The memory device has matrices A -1} , Q ^-1 , while the processor is mapping the frequency domain vector X to the time domain vector x using the ICZT.

,

,

,as well as

, And the system of claim 43, which does not store all the elements of ^P-1. 43. The system of claim 43, wherein the processor can perform its calculations with increased floating point precision. 48. The system of claim 48, wherein the processor can perform calculations on it with an increased floating point number greater than 32 bits. The memory device uses only O (n) of memory while the processor maps the frequency domain vector X to the time domain vector x using the ICZT, where n =. The system according to claim 43, which is max (M, N). ^{The processor performs O (n 2} ) or less basic operations while the processor maps the frequency domain vector X to the time domain vector x using the ICZT, where n = max. (M, N), the system according to claim 43. 51. The system of claim 51, wherein the processor performs an O (nlog (n)) operation during the mapping. The system includes a personal computer (PC), a system on chip (SoC), a system including a purpose-built integrated circuit (ASIC), a system including a field programmable gate array (FPGA), or a graphics processing unit (GPU). The system according to claim 43, which is at least one of the systems. 43. The system of claim 43, wherein the frequency domain vector X is not calculated from the particular time domain vector x using the chirp Z-transform with _{the same parameters A italics} and W _italics.

Images